Alzheimer's Disease Brain Phenotypes are Age-dependent

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Neurodegenerative Diseases

Background:

• The brain-age gap (BAG) biomarker assumes Alzheimer's Disease (AD) detection is age-invariant.
• This assumption is challenged by the widespread view of neurodegenerative disorders as accelerated aging.

Purpose of the Study:

• To investigate the necessity of age information in brain biomarkers for AD detection.
• To evaluate the performance of age-aware versus age-invariant brain MRI representations for AD identification.
• To explore the relationship between healthy aging and AD trajectories in the brain.

Main Methods:

• Invariant representation learning on brain MRI from 44,178 individuals.
• Development of age-aware and age-invariant neural representations.
• Causal analysis using conditional decoders and representational similarity analysis.

Main Results:

• Age-aware representations significantly outperformed age-invariant ones in AD detection (0.84 vs. 0.77 AUC).
• Healthy aging and AD diverge along multiple anatomical dimensions, with AD showing pathological temporal shifts and relative frontoparietal preservation.
• Age information is causally necessary for accurate AD detection, refuting the age-invariant assumption of BAG.

Conclusions:

• The brain-age gap (BAG) biomarker's conceptual flaw lies in discarding essential age-related information.
• Accurate AD detection requires biomarkers that preserve age-related brain structure, not age-independent ones.
• Multidimensional models capturing aging trajectories are superior to unidimensional summaries like BAG for understanding AD.